Cosmos: A Position-Resolution Causal Model for Direct and Indirect Effects in Protein Functions

Area of Science:

• Molecular Biology
• Systems Biology
• Computational Biology

Background:

• Deep mutational scanning (DMS) reveals protein variant effects on multiple molecular functions.
• Interpreting variant effects is complex when phenotypes are linked in molecular pathways, as downstream effects can be direct or indirect.

Purpose of the Study:

• To introduce Cosmos, a Bayesian framework for causal inference in multi-phenotype DMS data.
• To enable residue-level causal inference and distinguish direct from indirect functional effects of protein variants.
• To provide counterfactual interpretation by predicting downstream phenotypes after normalizing upstream ones.

Main Methods:

• Cosmos employs a Bayesian framework for residue-level causal inference.
• It uses position-level aggregation and Bayesian model selection to infer causal structures.
• The framework does not require phenotype-specific biophysical assumptions.

Main Results:

• Cosmos effectively distinguishes direct from indirect functional effects in multi-phenotype DMS data.
• Application to Kir2.1, PSD95-PDZ3, and KRAS datasets demonstrated its utility.
• The framework successfully inferred causal relationships and enabled counterfactual predictions.

Conclusions:

• Cosmos offers a generalizable and interpretable approach for disentangling causal relationships in high-throughput protein functional screens.
• It enhances the understanding of how protein variants impact molecular pathways.
• This framework advances the analysis of complex biological data from DMS experiments.